First published on O’Reilly’s OpenP2P on December 26, 2003.

This fall, I hosted a two-day brainstorming session for 30 or so people on the subject of social software. The event, sponsored by Cap-Gemini’s Center for Business Innovation and Nokia’s Insight and Foresight unit, took place in an open loft, and in addition to the usual “sit around a big table and talk to each other” format, we set up an in-room chat channel accessible over the WiFi network. We hosted the chat using Greg Elin’s modifications to Manuel Kiessling’s lovely ARSC (A Really Simple Chat) software. (Greg and I had used a similar setup in a somewhat different setting, and we were determined to experiment further at the social software event.)

The in-room chat created a two-channel experience — a live conversation in the room, and an overlapping real-time text conversation. The experiment was a strong net positive for the group. Most social software is designed as a replacement for face-to-face meetings, but the spread of permanet (connectivity like air) provides opportunities for social software to be used by groups who are already gathered in the same location. For us, the chat served as a kind of social whiteboard. In this note, I want to detail what worked and why, what the limitations and downsides of in-room chat were, and point out possible future avenues for exploration.

The Setup

The setup was quite simple. We were working in a large open loft, seated around a ring of tables, and we connected a WiFi hub to the room’s cable modem. Most of the participants had WiFi-capable laptops, and ARSC works in a browser, so there were no client issues. We put a large plasma screen at one end of the room.

We created a chat room for the event, and asked the participants using the chat to log in using their first name. In addition, we created a special username, Display, which we logged into a machine connected to the plasma screen. The Display interface had no text-entry field, suppressed control messages, and had its font set very large. This maximized screen real estate for user-entered messages, and made them readable even by participants sitting 10 meters away (though it minimized the amount of scroll-back visible on the plasma screen.)

Figure 1. The conference room.

Figure 2. Private and public screens.

Photos courtesy http://www.heiferman.com

We made the participants aware of the chat room at the beginning of the event, and set no other rules for its use (though at one point, we asked that people only use the chat room, saying nothing out loud for half an hour.) The chat room was available throughout the meeting. The first 10 minutes of the chat were the usual set of test messages, hellos, and other “My hovercraft is full of eels” randomness, but once the meeting got rolling, the chat room became an invaluable tool.

The Advantages

The chat room created several advantages.

1. It changed interrupt logic

Group conversations are exercises in managing interruptions. When someone is speaking, the listeners are often balancing the pressure to be polite with a desire to interrupt, whether to add material, correct or contradict the speaker, or introduce an entirely new theme. These interruptions are often tangential, and can lead to still more interruptions or follow-up comments by still other listeners. Furthermore, conversations that proceed by interruption are governed by the people best at interrupting. People who are shy, polite, or like to take a moment to compose their thoughts before speaking are at a disadvantage.

Even with these downsides, however, the tangents can be quite valuable, so if an absolute “no interrupt” rule were enforced, at least some material of general interest would be lost, and the frustration level among the participants consigned solely to passive listening would rise considerably. 

The chat room undid these effects, because participants could add to the conversation without interrupting, and the group could pursue tangential material in the chat room while listening in the real room. It was remarkable how much easier it was for the speaker to finish a complex thought without being cut off. And because chat participants had no way of interrupting one another in the chat room, even people not given to speaking out loud could participate. Indeed, one of our most active participants contributed a considerable amount of high-quality observation and annotation while saying almost nothing out loud for two days.

2. “Note to self” became “Note to world”

The more successful a meeting, the more “note to self” moments happen, where a light goes off in someone’s head, and they are moved to write the insight down for later examination. The chat channel provided an interesting alternative to personal note-taking, which was group note-taking. By entering “notes to self” into the chat, participants could both archive thoughts (the chat was logged) and to share those thoughts with the rest of the room to see what reactions they might spark. This is slightly different than simply altering interrupt logic, and more along the lines of Cory Doctorow’s “outboard brain” idea, because in this case, the chat was capturing material that would not otherwise have been shared with the group.

3. High-quality text annotation

What the spoken word has in emotive quality, it lacks in precision. Much interesting material thrown out during the course of group conversations is difficult to capture in an ideal form. When taking notes, it’s easy to misspell a name or mis-punctuate an URL, and things committed solely to memory can be difficult to retrieve later (“Somebody said something about a researcher in Oregon? Uraguay? The name began with a G …”). Comments in the chat log solved these problems — if the attendees were talking about Gerd Korteum’s work, or the Kuro5hin website, the spelling and punctuation were unambiguous.

There were two additional effects that improved the quality of the text annotation. Because everyone was connected to the Web, not just the local chat, the participants could Google for Web sites and quotes before they posted. (At one point during the Friday session, a fierce rain started, and someone pasted the US Weather service advisory for the area into the chat.) And because ARSC turns URLs into links, the rest of the group could click on a link in the chat window when it was added, so that new material could be glanced at and bookmarked in context, rather than hours or days later.

The annotation was also affected by the one-way relation between the real world conversation and the chat. Though it’s too early to know whether this was a bug or a feature, themes from the real world conversation were constantly reflected in the chat room, but almost never vice-versa. This suggests that the participants regarded the chat as a place for ancillary comments, rather than a separate but equal conversational space.

4. Less whispering, more \whispering

People whisper to one another during conferences, sometimes for good reasons (“What does UDDI mean?”) and sometimes for not-so-good ones (“So this Estonian guy goes into a bar …”). Like interrupting, however, a blanket “No whispering” ban would throw the good out with the bad, and would reduce the quality of the experience for the attendees. Furthermore, even when there is a good reason to whisper to someone, the larger the conference, the likelier it is you won’t be seated next to them.

The \whisper command in ARSC means that, topologically, everyone is seated next to everyone else. By typing “\whisper Rusty,” a participant could send a point-to-point message to Rusty without disrupting the meeting. Though the whispers weren’t logged, an informal poll at the end of the second day showed that a large majority of chat room participants had used \whisper at some point.

Ironically, the effectiveness of the \whisper command was somewhat limited by the “split screen” focus between the room and the chat. Because \whisper requests went to the invitee’s laptop, if someone was looking away from their screen for a few minutes, they would miss the invitation, since \whisper requests didn’t go to the plasma screen. One user suggested the addition of a \pssst function of some sort to get someone’s attention. Another possibility would be making a second \whisper-only window, so that \whisper conversations could be more asynchronous.

5. Alleviated boredom

Groups of people have diverse interests, so no matter how generally scintillating a meeting overall, at some point someone is going to find the subject at hand dull. The in-room chat helped alleviate this boredom, while keeping the participants talking to one another about the subject at hand.

This is the advantage hardest to understand in the abstract. When I talk about the in-room chat, people often ask “But isn’t that distracting? Don’t you want to make people pay attention to the speaker?” This is similar to the question from the early days of the Web: “But why have any outside links at all? Don’t you want to make people stay on your site?”

Once you assume permanet, whether from Wifi, Richochet, or GPRS, this logic crumbles. Anyone with a laptop or phone can, if they are bored, turn to the Internet, and the question becomes “Given that attendees will be using the network, would you rather have them talking to one another, or reading Slashdot?” The people who gathered in NYC came to converse with one another, and the in-room chat provided a way for them to meet that goal even when they were not riveted by the main event.

The Context

Chat as a meeting tool isn’t a universally good idea, of course. Every successful use of social software has environmental factors working in its favor.

First and foremost, the attendees were tech-savvy people who travel with WiFi-capable laptops and think about novel uses of social software, so they were inclined to want to use something like ARSC, even if only as an experiment. There was no resistance to trying something so odd and unproven, as there might be in less-techie groups.

The group was also self-motivated. Because their attendance was optional, they largely stayed on-topic. One can easily imagine that in a meeting where attendance is passive and forced (“The boss wants everyone in the conference room at 5:45”) the contents of the chat would be much more tangential (to say nothing of libelous). Since most parliamentary rules, whether formal or informal, begin with the premise that only one person can speak at once, and then arrange elaborate rules for determining who can speak when, the presence of an alternate channel could severely disrupt highly-structured meetings, such as client conferences or legal negotiations. Whether this would be a bug or a feature depends on your point of view.

The goals of the meeting were in synch with the experience the chat room offered. We were not trying to forge a working group, get to consensus, or even converge on a small set of ideas. Indeed, the goals of the meeting were explicitly divergent, trying to uncover and share as much new material as possible in a short period. The chat room aided this goal admirably.

The scale of the meeting also worked in our favor. The group was large enough that sitting around a table with a laptop open wasn’t rude or disruptive, but small enough that everyone could use a single chat room. At one point during the Saturday session, we broke into small groups to brainstorm around specific problems, and though there was no explicit request to do so, every single member of the group shut their laptop screens for two hours. Groups of six are small enough that all the members can feel engaged with the group, and the chat would have been much less useful and much more rude in that setting.

On the other hand, whenever things got really active on the chat channel (we averaged about four posts a minute, but it sometimes spiked to 10 or so), people complained about the lack of threading, suggesting that 30 was at or near an upper limit for participation.

Meeting Structure

There were also some more technical or formal aspects of the meeting that worked in our favor.

The plasma screen showing the Display view was surprisingly important. We had not announced the WiFi network or chat channel in advance, and we had no idea how many people would bring WiFi-capable laptops. (As it turned out, most did.) The plasma screen was there to share the chat room’s contents with the disconnected members. However, the screen also added an aspect of social control — because anything said in the chat room was displayed openly, it helped keep the conversation on-topic. Curiously, this seemed to be true even though most of the room was reading the contents of the chat on their laptop screens. The plasma screen created a public feeling without actually exposing the contents to a “public” different from the attendees.

During a brief post-mortem, several users reported using the plasma screen for chunking, so that they could mainly pay attention to the speaker, but flash their eyes to the screen occasionally to see what was in the chat room, taking advantage of the fact that most people read much faster than most speakers talk. (Viz. the horror of the speaker who puts up a PowerPoint page, and then reads each point.)

There were two bits of organizational structure that also helped shape the meeting. The first was our adoption of Jerry Michalski’s marvelous “Red Card/Green Card” system, where participants were given a set of colored cards about 20 cm square in three colors, red, green, and gray. The cards were used to make explicit but non-verbal commentary on what was being said at the time. A green card indicates strong assent, red strong dissent, and gray confusion.

In an earlier experiment with ARSC, Greg added virtual cards; users could click on red or green icons and have those added to the chat. This proved unsatisfying, and for this meeting we went back to the use of physical cards. The use of the cards to indicate non-verbal and emotive reactions seemed to provide a nice balance with the verbal and less emotive written material. At one point, we spent half an hour in conversation with the only rule being “No talking.” The entire room was chatting for 30 minutes, and even in that situation, people would physically wave green cards whenever anyone posted anything particularly worthy in the chat room.

While the no-talking experiment was interesting, it was not particularly useful. One of the participants whose work was being discussed in the chat (he had just finished talking when we entered the no-talking period) reported missing the actual verbal feedback from colleagues. The chat comments made about his ideas, while cogent, lacked the emotional resonance that makes face-to-face meetings work. By enforcing the no-talking rule, we had re-created some of the disadvantages of virtual meetings in a real room.

The other bit of organizational structure was borrowed from Elliott Maxwell and the Aspen Institute, where participants wanting to speak would turn their name cards vertically, thus putting comments in a queue. This was frustrating for many of the participants, who had to wait several minutes to react to something. This also severely altered interrupt logic. (At several points, people gave up their turn to speak, saying “the moment has passed.”) Despite the frustration it caused, this system kept us uncovering new material as opposed to going down rat holes, and it made the chat room an attractive alternative for saying something immediate.

The Disadvantages

Though we found ARSC to be a useful addition to the meeting, there was an unusually good fit between the tool and the environment. For every favorable bit of context listed above, there is a situation where an in-room chat channel would be inappropriate. Meetings where the attention to the speaker needs to be more total would be problematic, as would situations where a majority of the audience is either not connected or uncomfortable with chat.

Even in this group, not everyone had a laptop, and for those people, the chat contents were simply a second channel of information that they could observe but not affect. Absolute ubiquity of the necessary hardware is some way off for even tech-savvy groups, and several years away, at least, for the average group. Any meeting wanting to implement a system like this will have to take steps to make the chat optional, or to provide the necessary hardware where it is lacking.

It may also be that increasing phone/PDA fusion will actually reduce the number of laptops present at meetings. Using social tools at events where phones are the personal device of choice will require significant additional thought around the issues of small screens, thumb keyboards, and other ergonomic issues.

In-room chat is unlikely to be useful for small groups (fewer than a dozen, at a guess), and its usefulness for groups larger than 30 may also be marginal (though in that case, ways of providing multiple chat rooms may be helpful).

Given that the most profound effects of the chat were in changing interrupt logic, many of the downsides came from the loss of interruption as a tool. As annoying as interruptions may be, they help keep people honest, by placing a premium on brevity, and on not spinning castles in the air. Without interruption, the speaker loses an important feedback mechanism, and may tend towards long-winded and unsupported claims. At the very least, the use of in-room chat puts a premium on strong moderation of some sort, to make up for the structural loss of interruption.

Perhaps most importantly, it will almost certainly be unhelpful for groups that need to function as a team. Because the two-track structure encourages a maximum number of new and tangential items being placed together, it would probably be actively destructive for groups where consensus was a goal. As Steven Johnson has noted about the event, the chat room moved most of the humor from real world interjections to network ones, which preserved the humor but suppressed the laughter. (Most of then time when people write “lol,” they aren’t.) Though this helps on the “interrupt logic” front, it also detracts from building group cohesion.

This experiment was relatively small and short, having been applied to one group over two days. It would be interesting to know what the effect would be for groups meeting for longer periods. On the first day, we averaged not quite three and a half posts a minute, occasionally spiking to nine or 10. On the second day, the average rose to just over four posts a minute, but the spikes didn’t change, suggesting that users were becoming more accustomed to a steady pace of posting.

During the half-hour of chat only/no talking out loud, the average nearly tripled, to over 11 posts a minute, and occasionally spiking to 18, suggesting that during the normal sessions, users were paying what Linda Stone calls “continuous partial attention” to both the chat room and the real room, and with the structures of the real room artificially suppressed, the chat room exploded. (The question of whether the change in posting rate was uniform among all users is difficult to answer with the small sample data.)

Several additional experiments suggest themselves: 

• For conferences whose sessions average no more than 30 users, having each room assigned its own chat channel could let users listening to the same talk find one another with little effort. Likewise, finding ways of forking large groups into multiple chats might be worthwhile, whether using specific characteristics (UI designers in one, information architects in another, and so on) or arbitrary ones (even or odd date of birth) to keep the population of any one channel in the 12-25 range. (The lack of plasma screen as a public mediating factor might be an issue.)
• ARSC translates URLs into clickable links. This suggests other regular expressions that could be added. “A:” plus an author name or “T:” plus a title could be turned into Amazon lookups. A QuoteBot might be very useful, as several times during the two days someone asked “Who said ‘Let the wild rumpus start’?” A social network analysis bot might be interesting, logging things like most and least frequent posters, and social clustering, and reflecting those back to the group. (Cameron Marlow of blogdex wrote a simple program during the meeting to display the number of chat posts per user.)
• The social network angle, of course, is hampered by the lack of threading in chat. This is obviously a hard problem, but several people wondered whether there might be a lightweight way to indicate who you are responding to, to create rudimentary threading. This may be a problem best fixed socially. If we had asked people to adopt the general IRC convention of posting with the name of the recipient first (“greg: interesting idea, but almost certainly illegal”), we might have gotten much better implicit threading. This in turn would have been greatly helped by tab-completion of nicknames in ARSC.
• The \whisper function is secret, rather than private. In a real meeting, seeing who is having a side conversation can allow the group as a whole to feel the overall dynamic, so a private \whisper function might be an interesting addition, entering lines into the chat like “Clay whispers to Cameron,” but providing no information about the content of those conversations.
• Likewise, it might be useful to flag interesting or relevant posts for later review. If someone says something particularly cogent, other users could click a link next to the post labeled “Archive me,” and in addition to appearing in the general log, such posts would go to a second “flagged comments only” log.
• Greg provided a polling function, but you had to click off the chat page to get there, and it was only used once, as a test. Given this failure, polling and voting functions may need to appear directly in the chat room to be useful. Bots are an obvious interface to do this as well. A PollingBot could ask questions in the chat room and accept answers by \whisper.
• Given that meetings generally involve people looking at similar issues from different backgrounds, DCC-like user-to-user file transfer might be a valuable tool for sharing background materials among the participants, by letting them send local files as well as URLs over the chat interface.
• We got close to the edge of IRC-style chaos, where the chat scrolls by too fast to read it. A buffering chat channel might solve this problem, by having some maximum rate set on the order of 120-150 words a minute, and then simply delaying posts that go over that limit into the next minute, and so on. This congestion queuing would let everyone say what they want to say without dampening the ability of other participants to take it all in before reaction. 
• Finally, ARSC is server-based. With Zeroconf networking, it might be possible to set up ad hoc, peer-to-peer networks of laptops without needing to coordinate anything in advance. Likewise, while DCC-ish file transfer might be valuable for person-to-person file sharing, the ability to post “I have a draft of my article on my hard drive at such and such a local address” and have that material be as accessible as if it were on the Web would make public sharing of background materials much easier.

Conclusion

Real world groups are accustomed to having tools to help them get their work done — flipcharts, white boards, projectors, and so on. These tools are typically used by only one person at a time. This experiment demonstrated the possibility of social tools, tools that likewise aid a real-world group, but that are equally accessible to all members at the same time. There are a number of other experiments one can imagine, from using the chat to accept input from remote locations to integrating additional I/O devices such as cameras or projectors. The core observation, though, is that under certain conditions, groups can find value in participating in two simultaneous conversation spaces, one real and one virtual.

First published December 17, 2003 on the “Networks, Economics, and Culture” mailing list. 

For years, the US Government has been terrified of losing surveillance powers over digital communications generally, and one of their biggest fears has been broad public adoption of encryption. If the average user were to routinely encrypt their email, files, and instant messages, whole swaths of public communication currently available to law enforcement with a simple subpoena (at most) would become either unreadable, or readable only at huge expense. 

The first broad attempt by the Government to deflect general adoption of encryption came 10 years ago, in the form of the Clipper Chip. The Clipper Chip was part of a proposal for a secure digital phone that would only work if the encryption keys were held in such a way that the Government could get to them. With a pair of Clipper phones, users could make phone calls secure from everyone except the Government. 

Though opposition to Clipper by civil liberties groups was swift and extreme, the thing that killed it was work by Matt Blaze, a Bell Labs security researcher, showing that the phone’s wiretap capabilities could be easily defeated, allowing Clipper users to make calls that even the Government couldn’t decrypt. (Ironically, ATT had designed the phones originally, and had a contract to sell them before Blaze sunk the project.)

The Government’s failure to get the Clipper implemented came at a heady time for advocates of digital privacy — the NSA was losing control of cryptographic products, Phil Zimmerman had launched his Pretty Good Privacy (PGP) email program, and the Cypherpunks, a merry band of crypto-loving civil libertarians, were on the cover of the second issue of Wired. The floodgates were opening, leading to…

…pretty much nothing. Even after the death of Clipper and the launch of PGP, the Government discovered that for the most part, users didn’t want to encrypt their communications. The most effective barrier to the spread of encryption has turned out to be not control but apathy. Though business users encrypt sensitive data to hide it from one another, the use of encryption to hide private communications from the Government has been limited mainly to techno-libertarians and a small criminal class.

The reason for this is the obvious one: the average user has little to hide, and so hides little. As a result, 10 years on, e-mail is still sent as plain text, files are almost universally unsecured, and so on. The Cypherpunk fantasy of a culture that routinely hides both legal and illegal activities from the state has been defeated by a giant distributed veto. Until now. 

It may be time to dust off that old issue of Wired, because the RIAA is succeeding where 10 years of hectoring by the Cypherpunks failed. When shutting down Napster turned out to have all the containing effects of stomping on a tube of toothpaste, the RIAA switched to suing users directly. This strategy has worked much better than shutting down Napster did, convincing many users to stop using public file sharing systems, and to delete MP3s from their hard drives. However, to sue users, they had to serve a subpoena, and to do that, they had to get their identities from the user’s internet service providers.

Identifying those users has had a second effect, and that’s to create a real-world version of the scenario that drove the invention of user-controlled encryption in the first place. Whitfield Diffie, inventor of public key encryption, the strategy that underlies most of today’s cryptographic products, saw the problem as a version of “Who will guard the guardians?” 

In any system where a user’s identity is in the hands of a third party, that third party cannot be trusted. No matter who the third party is, there will be at least hypothetical situations where the user does not want his or her identity revealed, but the third party chooses or is forced to disclose it anyway. (The first large scale example of this happening was the compromise of anon.penet.fi, the anonymous email service, in 1994.) Seeing that this problem was endemic to all systems where third parties had access to a user’s identity, Diffie set out to design a system that put control of anonymity directly in the hands of the user.

Diffie published theoretical work on public key encryption in 1975, and by the early 90s, practical implementations were being offered to the users. However, the scenario Diffie envisioned had little obvious relevance to users, who were fairly anonymous on the internet already. Instead of worrying now about possible future dangers, most users’ privacy concerns centered on issues local to the PC, like hiding downloaded pornography, rather than on encrypting network traffic.

However, Diffie’s scenario, where legal intervention destroys the users’ de facto privacy wherever it is in the hands of commercial entities, is now real. The RIAA’s successful extraction of user identity from internet service providers makes it vividly clear that the veil of privacy enjoyed by the average internet user is diaphanous at best, and that the obstacles to piercing that veil are much much lower than for, say, allowing the police to search your home or read your (physical) mail. Diffie’s hypothetical problem is today’s reality. As a result, after years of apathy, his proposed solution is being adopted as well.

In response to the RIAA’s suits, users who want to share music files are adopting tools like WINW and BadBlue, that allow them to create encrypted spaces where they can share files and converse with one another. As a result, all their communications in these spaces, even messages with no more commercial content than “BRITN3Y SUX!!!1!” are hidden from prying eyes. This is not because such messages are sensitive, but rather because once a user starts encrypting messages and files, it’s often easier to encrypt everything than to pick and choose. Note that the broadening adoption of encryption is not because users have become libertarians, but because they have become criminals; to a first approximation, every PC owner under the age of 35 is now a felon. 

The obvious parallel here is with Prohibition. By making it unconstitutional for an adult to have a drink in their own home, Prohibition created a cat and mouse game between law enforcement and millions of citizens engaged in an activity that was illegal but popular. As with file sharing, the essence of the game was hidden transactions — you needed to be able to get into a speakeasy or buy bootleg without being seen.

This requirement in turn created several long-term effects in American society, everything from greatly increased skepticism of Government-mandated morality to broad support for anyone who could arrange for hidden transactions, including organized crime. Reversing the cause did not reverse the effects; both the heightened skepticism and the increased power of organized crime lasted decades after Prohibition itself was reversed.

As with Prohibition, so with file sharing — the direct effects from the current conflict are going to be minor and over quickly, compared to the shifts in society as a whole. New entertainment technology goes from revolutionary to normal quite rapidly. There were dire predictions made by the silent movie orchestras’ union trying to kill talkies, or film executives trying to kill television, or television executives trying to kill the VCR. Once those technologies were in place, however, it was hard to remember what all the fuss was about. Though most of the writing about file sharing concentrates on the effects on the music industry, whatever new bargain is struck between musicians and listeners will almost certainly be unremarkable five years from now. The long-term effects of file sharing are elsewhere.

The music industry’s attempts to force digital data to behave like physical objects has had two profound effects, neither of them about music. The first is the progressive development of decentralized network models, loosely bundled together under the rubric of peer-to-peer. Though there were several version of such architectures as early as the mid-90s such as ICQ and SETI@Home, it took Napster to ignite general interest in this class of solutions.

And the second effect, of course, is the long-predicted and oft-delayed spread of encryption. The RIAA is succeeding where the Cypherpunks failed, convincing users to trade a broad but penetrable privacy for unbreakable anonymity under their personal control. In contrast to the Cypherpunks “eat your peas” approach, touting encryption as a first-order service users should work to embrace, encryption is now becoming a background feature of collaborative workspaces. Because encryption is becoming something that must run in the background, there is now an incentive to make its adoption as easy and transparent to the user as possible. It’s too early to say how widely casual encryption use will spread, but it isn’t too early to see that the shift is both profound and irreversible.

People will differ on the value of this change, depending on their feelings about privacy and their trust of the Government, but the effects of the increased use of encryption, and the subsequent difficulties for law enforcement in decrypting messages and files, will last far longer than the current transition to digital music delivery, and may in fact be the most important legacy of the current legal crackdown.